Online Signature Verification and Handwriting Classification

137 Pages·2017·10.27 MB·English

ONLINE SIGNATURE VERIFICATION
AND HANDWRITING CLASSIFICATION
Ph.D. dissertation
erika griechisch
dr jános csirik
Supervisor: .
Faculty of Science and Informatics
Doctoral School of Computer Science
Institute of Informatics, University of Szeged
2018
Szeged
Erika Griechisch: Online Signature Verification and Handwriting Classification, PhD
2018
in Computer Science, ©
PUBLICATIONS
Some ideas, tables and figures have appeared in the following publications:
2011
AFHA
ános sirik oltán ingl and rika riechisch
J C , Z G E G : The Effect of Training
Data Selection and Sampling Time Intervals on Signature Verification, Proceed-
ings of the First International Workshop on Automated Forensic Handwrit-
6 10
ing Analysis, pp. –
DOI: http://dx.doi.org/10.1.1.369.5581
2011
CIARP
orst unke ános sirik oltán ingl and rika riechisch
H B , J C , Z G E G : On-
line signature verification method based on the acceleration signals of handwriting
7042
samples, Lecture Notes in Computer Science, Volume , Progress in Pat-
tern Recognition, Image Analysis, Computer Vision, and Applications, pp.
499 506
–
URL: http://forums.graphonomics.org/showthread.php?t=194
2013
IGS
rika riechisch uhammad mran alik and arcus iwicki
E G , M I M M L : On-
line Signature Verification using Accelerometer and Gyroscope, Proceedings of
16 143 146
th International Graphonomics Society Conference, pp. – .
DOI: http://dx.doi.org/10.1109/ICDAR.2013.82
2013
ICDAR
rika riechisch uhammad mran alik and arcus iwicki
E G , M I M M L : On-
line Signature Analysis based on Accelerometric and Gyroscopic pens and Leg-
12
endre Series, Proceedings of th International Conference on Document
25 28 374
Analysis and Recognition, Washington, DC, USA, August - , pp. –
378
DOI: http://dx.doi.org/10.1109/ICDAR.2013.82
2014
ICFHR
rika riechisch uhammad mran alik and arcus iwicki
E G , M I M M L : On-
line Signature Verification Based on Kolmogorov-Smirnov Distribution Distance,
14
Proceedings of th International Conference on Frontiers in Handwriting
1 4 2014 738 742
Recognition, Crete, Greece, September - , , pp. –
DOI: http://dx.doi.org/10.1109/ICFHR.2014.129
2015
ICDAR
rika riechischand rika encsik
E G E B :HandednessDetectionofOnlineHand-
13
writings based on Horizontal Strokes, Proceedings of th International Con-
1272 1277
ference on Document Analysis and Recognition, pp. –
DOI: http://dx.doi.org/10.1109/ICDAR.2015.7333953
2017
IGS
rika encsik and rika riechisch
E B E G : The frequency of occurence of hand-
18
writing features in online male and female handwriting, Proceedings of th
169 172
International Graphonomics Society Conference pp. –
We have seen that computer programming is an art,
because it applies accumulated knowledge to the world,
because it requires skill and ingenuity, and especially
because it produces objects of beauty.
1
— Donald E. Knuth [ ]
ACKNOWLEDGMENTS
I am grateful to my supervisor János Csirik, who made it possible for me to
7
work in this interesting research topic and guided me in the last years. I am
also grateful to all my co-authors in this period: Inés Baldatti, Erika Bencsik,
Horst Bunke, Zoltán Gingl, Marcus Liwicki, Muhammad Imran Malik and last
but not least Gábor Németh. Thank you to David P. Curley who has improved
the English of my papers.
Many thanks to my parents who made it possible for me to come and study in
Szeged and together with my sister who supported me morally and emotionally
in my life. Special gratitude goes out to my beloved husband Gábor Németh,
who has also supported me along the long journey.
I am also grateful to the Department of Medical Physics and Informatics (Uni-
versity of Szeged) and the head of the institute Ferenc Bari who offered me
4
research position years ago and since then supported the completion of my
PhD dissertation with great patience.
In addition thank you to all my family members, friends, collegaues and stu-
dents who increased the size of my databases with signature and handwriting
samples.
Erika Griechisch
16 2018
th May

CONTENTS
history definitions and databases 1
i ,
1 introduction 3
11 3
. History of forensics...........................................................
111 3
. . Handwriting examination ..........................................
112 4
. . Differences between signature and handwriting................
12 5
. Handwriting analysis ........................................................
13 6
. Role of handwritten signature and handwriting ........................
14 6
. Acquisition for automatic examination....................................
141 7
. . Challenges.............................................................
15 7
. Related reviews and comprehensive studies .............................
2 automated online verification 9
21 10
. Acquisition.....................................................................
22 11
. Preprocessing..................................................................
221 11
. . Normalization.........................................................
222 11
. . Segmentation..........................................................
23 13
. Feature extraction.............................................................
231 13
. . Parametric approach.................................................
232 13
. . Functional approach .................................................
233 14
. . Feature analysis.......................................................
24 14
. Classification...................................................................
241 14
. . Type of forged signatures...........................................
242 15
. . Distance-based classification .......................................
243 15
. . Dynamic Time Warping (DTW)....................................
244 16
. . Hidden Markov Models.............................................
245 19
. . Artificial Neural Networks .........................................
246 20
. . Support Vector Machine.............................................
25 21
. Evaluation......................................................................
26 23
. Databases.......................................................................
261 2003 23
. . MCYT .............................................................
262 2004 23
. . SVC ................................................................
263 2009 24
. . SUSIG ( ) ..........................................................
264 2009 25
. . BSEC ..............................................................
265 2009 25
. . SigComp ..........................................................
266 2011 26
. . SigComp ..........................................................
267 2013 26
. . SigWiComp ......................................................
268 2015 29
. . SigWiComp ......................................................
269 4 2010 29
. . NSigComp ......................................................
2610 4 2012 30
. . NSigComp ......................................................
i
contents
ii
27 32
. Winning methods .............................................................
271 2004 32
. . SVC ................................................................
272 2009 32
. . BSEC ..............................................................
273 2011 33
. . ESRA’ .............................................................
274 2009 33
. . SigComp ..........................................................
275 2011 33
. . SigComp ..........................................................
276 2013 34
. . SigWiComp ......................................................
277 2015 34
. . SigWiComp ......................................................
online signature verification and comparison 37
ii
3 datasets 39
31 39
. AccSigDb .......................................................................
311 39
. . Acquisition device....................................................
312 1 1 40
. . Subset (AccSigDb ) ................................................
313 2 2 41
. . Subset (AccSigDb ) ................................................
32 41
. GyroSigDb .....................................................................
321 43
. . Angular momentum .................................................
33 43
. Overlap of writers.............................................................
4 online signature verification and classification 45
41 45
. Acceleration based signature verification.................................
411 45
. . Related work ..........................................................
412 1 46
. . Examination on the AccSigDb dataset...........................
413 46
. . Results..................................................................
414 48
. . Summary ..............................................................
42 49
. Different reference signature and time period ...........................
421 49
. . Selection of reference signatures...................................
422 50
. . Signatures from different time periods ...........................
423 53
. . Conclusion.............................................................
43 54
. Legendre approximation in online signature classification ............
431 54
. . Legendre approximation for feature extraction .................
432 55
. . SVM classifier for classification ....................................
433 55
. . Results..................................................................
434 55
. . Conclusion.............................................................
5 kolmogorov smirnov metric in signature verification 57
-
51 57
. The proposed algorithm .....................................................
511 58
. . Preprocessing .........................................................
512 58
. . Feature extraction ....................................................
513 58
. . Classification ..........................................................
514 63
. . Algorithm..............................................................
52 64
. Results ..........................................................................
53 66
. Conclusion .....................................................................
contents
iii
online handwriting classification 67
iii
6 handedness detection 69
61 69
. Related work...................................................................
62 70
. Methodology...................................................................
63 71
. Algorithm ......................................................................
631 73
. . Majority voting (MV) ................................................
632 76
. . Left-condition (LC ) .................................................
k
64 78
. Experiments and results .....................................................
65 82
. Conclusion and future plans................................................
7 gender detection 83
71 83
. Introduction....................................................................
72 84
. Related work...................................................................
73 86
. Forensic examination and results ..........................................
74 92
. Automatic examination and results........................................
75 93
. Conclusion .....................................................................
summary 95
iv
97
Summary in English
101
Összegzés magyarul
bibliography 105
LIST OF FIGURES
11 4
. Biometrical traits..............................................................
12 5
. Stereo microscope and an ESDA ..........................................
13 7
. Signature of Louis XIV of France ..........................................
14 2009 8
. A signature sample taken from SigComp database...............
15 8
. Online handwriting sample from IAM-OnDB...........................
21 9
. Steps of a signature verification method..................................
22 3 12
. Wacom Intuos graphics tablets ...........................................
23 12
. Sigma, Omega, Gamma and Alpha Signature Pads from signotec ..
24 12
. Anoto pen......................................................................
25 17
. Visualization of DTW........................................................
26 4 2 18
. An HMM with states and possible observations ...................
27 19
. A Neural Network with one hidden layer ...............................
28
. SVMappliesakernelfunctiontodetermineanon-linearseparator
21
for the classes as a linear separator in a higher dimension............
29 2004 24
. Online signature sample taken from the SVC database ..........
210 2009 2 27
. Sample taken from BSEC’ DS database............................
211 2009 3 27
. Sample taken from the BSEC’ DS database .......................
212 2009 27
. Online signature sample taken from the SigComp database ....
213 2013 28
. Samples taken from the SigWiComp Japanese online database.
31 40
. The accelerometer is mounted close to the tip of the pen .............
32 41
. Block diagram of the data acquisition system ...........................
33
. The images and acceleration signals of two genuine signatures
1 42
and one forged signature from AccSigDb ...............................
34
. The images and corresponding acceleration signals of two gen-
1
uine signatures and two forged signatures from AccSigDb and
2 42
AccSigDb .....................................................................
35 43
. The gyroscope is mounted to the pen.....................................
36
. Four genuine signatures written by the same writer from dataset
44
AccSigDb.......................................................................
37
. Two genuine signatures and the corresponding signals from wri-
ter NG (GyroSig, left/right: x/y axis) .................................... 44
41
. False rejection and false acceptance rates depending on the con-
47
stant multiplier................................................................
42 53
. False acceptance and false rejection rates ................................
43 56
. Accuracy of the classification...............................................
51
. Plotsoftwoempiricalcumulativedistributionfunctions,theblack
arrow shows the maximal difference between the functions, thus
59
the two-sample KS distance ................................................
iv